PhD Researcher: Effective Models of Neural Network Training

This PhD position at Inria’s MEGAVOLT team in Paris focuses on creating effective models to understand the optimization and statistical performance of neural network training. The successful candidate will derive theoretical models incorporating key phenomena in neural networks, validate them empirically through simulations, and demonstrate their accuracy. The work will involve close collaboration with physicists at EPFL (Lausanne), combining computational and theoretical approaches to advance the field of machine learning theory. The role is ideal for candidates with a strong background in theoretical computer science, physics, or applied mathematics, seeking to contribute to foundational research in AI.

• Conduct bibliographic research to identify key phenomena in neural network training and optimization.
• Develop effective theoretical models that capture essential dynamics of neural network training.
• Perform numerical simulations to empirically validate the proposed models.
• Collaborate with physicists at EPFL to integrate theoretical insights with empirical findings.
• Write and publish scientific articles detailing research findings and model validation.
• Engage in scientific outreach and contribute to the MEGAVOLT team’s working group activities.